Solaris 10 amd64 - pollsys loop of a java programme
Hi!
I've got a strange problem: my java command line aplication that connects to a socket (where another java apl is listening on) does its job (currently transferring an object throgh connection, read it and returns output to STDOUT) and after that it falls into a kind of pollsys loop:
truss -p 23312
/2: pollsys(0x00000000, 0, 0xFBC8BE58, 0x00000000) = 0
/2: pollsys(0x00000000, 0, 0xFBC8BE58, 0x00000000) = 0
(...)
It goes on with this syscall all the time and after a while (several minutes(!) to hours times differ) it ends OK. The process receives signals but thoesn't handle 'em uncluding SIGKILL (sic! never seen on Sol before). What can stop kernell from killing crazy process?
^C/1: Received signal #2, SIGINT, in lwp_cond_wait() [caught]
It's a widely testes and used production aplication that we had never such probelms with under Sparc Solaris9 before migrating (test period) to the amd64 Opteron 244 TYAN platform with Solaris10 x86. Well, nothing but pain so far.
Can anybody give me a hint what to do with it then? Is it likely to be a hardware problem? How to track it further, maybe someone could suggest some DTrace solution?